JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR

HUMAN OBSER VING: MAINTAINED BY NEGA TIVEINFORMATIVE STIMULI ONLY IF CORRELATED WITH

IMPROVEMENT IN RESPONSE EFFICIENCYDAVID A. CASE, EDMUND FANTINO, AND JOHN WIXTED

UNIVERSITY OF CALIFORNIA, SAN DIEGO

Two experiments investigated the effect of observing responses that enabled collegestudents to emit more efficient distributions of reinforced responses. In Experiment 1, thegains of response efficiency enabled by observing were minimized through use of identicallow-effort response requirements in two alternating variable-interval schedules. Thesecomprised a mixed schedule of reinforcement; they differed in the number of money-backed points per reinforcer. In each of three choices between two stimuli that varied intheir correlation with the variable-interval schedules, the results showed that subjectspreferred stimuli that were correlated with the larger average amount of reinforcement.This is consistent with a conditioned-reinforcement hypothesis. Negative informativestimuli - that is, stimuli correlated with the smaller of two rewards- did not maintain asmuch observing as stimuli that were uncorrelated with amount of reward. In Experiment2, savings in effort made possible by producing S- were varied within subjects by alter-nately removing and reinstating the response-reinforcement contingency in a mixedvariable-interval/extinction schedule of reinforcement. Preference for an uncorrelatedstimulus compared to a negative informative stimulus (S-) decreased for each of six sub-jects, and usually reversed when observing permitted a more efficient temporal distribu-tion of the responses required for reinforcement; in this case, the responses were pulls on arelatively high-effort plunger. When observing the S- could not improve response effi-ciency, subjects again chose the control stimulus. All of these results were inconsistent withthe uncertainty-reduction hypothesis.

Key words: conditioned reinforcement, observing, choice, information, response effi-ciency, lever pressing, adults

Observing responses are those that mayproduce stimuli correlated with the com-ponents of a mixed schedule of reinforcement.In a mixed schedule of reinforcement, theschedules of reinforcement that operate in-dependently in each component all share thesame stimulus. Other than identifying thecomponent currently operative, by effectivelyconverting the mixed to a multiple schedule,observing responses have no effect on primaryreinforcement (Wyckoff, 1952).A large number of experiments investi-

gating why the contingent stimuli reinforceobserving have produced results consistentThis research was supported by NSF Grant BNS

83-02963 and by NIMH Grant MH-20752 to the Uni-versity of California at San Diego. John Wixted is nowat Emory University. Reprints may be obtained fromeither David A. Case or Edmund Fantino, Depart-ment of Psychology C-009, University of California,San Diego, La Jolla, California 92093.

with the conditioned-reinforcement hypothesis(see Fantino, 1977; Fantino & Case, 1983).This hypothesis proposes that the stimuli arereinforcing because of their positive correlationwith primary reinforcement. Although exten-sive support has been found in several studiesusing pigeons (Bowe & Dinsmoor, 1983; Case& Fantino, 1981; Dinsmoor, Browne, &Lawrence, 1972; Killeen, Wald, & Cheney,1980; Mulvaney, Dinsmoor, Jwaideh, &Hughes, 1974; and others reviewed by Fan-tino, 1977), controversy surrounds studiesconducted with adult humans (Fantino &Case, 1983; Perone & Baron, 1980). In par-ticular, Perone and Baron reported data thatwere apparently inconsistent with the con-ditioned-reinforcement hypothesis, becausethey showed that observing could be reinforcedby a negative discriminative stimulus (S-). Inlight of the evidence on pigeon observing,

289

1985, 43, 289-300 NUMBER 3 (MAY)

DAVID A. CASE, EDMUND FANTINO, and JOHN WIXTED

Perone and Baron accounted for the dis-crepancy by proposing that for organismshigher on the phylogenetic scale (e.g., pri-mates) the information or uncertainty-re-duction provided by a stimulus (i.e., tellingsubjects that extinction is currently in effect) isa more powerful determinant of observingthan is conditioned reinforcement (Schrier,Thompson, & Spector, 1980; see also Berlyne,1960; Bloomfield, 1972). The results of Fan-tino and Case (1983) challenged this inter-pretation, however, for these authors found noevidence of reinforcement by uncertainty re-duction in their study of adult observing. Fan-tino and Case proposed a resolution of thefindings based upon a potentially importantdifference. Subjects produced reinforcement inthe Perone and Baron experiments by pullinga plunger (or plungers, depending on the ex-periment) that required minimum forces offrom 5 to 20 ft-lbs to operate, which is approx-imately 25 to 100 times the minimum force re-quired to operate the observing levers. In con-trast, reinforcement occurred independent ofany responding in the experiments of Fantinoand Case. In the Perone and Baron study, thepossibility that observing was reinforced by S-because it reduced uncertainty is confoundedwith the possibility that observing S- permit-ted a more efficient distribution of the con-siderably more effortful responding. Specifi-cally, subjects could reduce ineffective re-sponding by ceasing to pull the plunger afterproducing S-. Thus, the increased responseefficiency made possible by observing S-could have masked any effect of its negativecorrelation with reinforcement (a possibilityconsidered, but not favored, by Perone andBaron). The response-independent reinforce-ment procedure used by Fantino and Caseeliminated the possiblitiy that observing couldaffect the rate or distribution of either rein-forcement or responding required for rein-forcement. In this case, S- failed to reinforceobserving.The present report describes two experi-

ments with human adults, testing the inter-pretation of Fantino and Case (1983). Observ-ing maintained by S-, a less positivediscriminative stimulus, was measured in dif-

ferent conditions in which observing eithercould or could not be correlated with a moreefficient distribution of responses required forreinforcement. Experiment 1 used response-dependent reinforcement, as in Perone andBaron's (1980) research, but this time con-tingencies were arranged so as to minimize theimprovement in response efficiency permittedthrough observing a less positive discrim-inative stimulus. Experiment 2 closely dup-licated response contingencies employed byFantino and Case in one condition, and thoseemployed by Perone and Baron in another, inorder to vary within the same subjects theresponse efficiency made possible by observingS-. The critical comparison throughout iswhether subjects prefer observing S- over astimulus that is uncorrelated with reinforcerrate or amount (SU). The uncertainty-reduction hypothesis predicts preference forthe former stimuli because only they are infor-mative about reinforcement; the uncorrelatedstimulus is by definition uninformative.Theconditioned-reinforcement hypothesis predictspreference for the uncorrelated stimulus, pro-vided observing does not permit substantiallymore efficient responding, because Su has amore positive correlation with reinforcement.

EXPERIMENT 1

The explanation proposed by Fantino andCase (1983) in accounting for Perone andBaron's (1980) results depends upon there be-ing differences between the studies with respectto response efficiency permitted throughobserving S- (less positive discriminativestimuli). Fantino and Case elected to useresponse-independent reinforcement in orderto eliminate all possible considerations of re-sponse efficiency in accounting for their re-sults. Response-independent reinforcement perse was assumed not to be critical except for con-siderations of response efficiency. The presentexperiment tested this assumption, and thegenerality of the results of Fantino and Case,by using a response-dependent reinforcementprocedure in which observing enabled only aminimal saving of effort in making the rein-forced response. Specifically, a mixed schedule

290

RESPONSE EFFICIENCY AND OBSER VING

was used; it was comprised of two identicalvariable-interval 30-s (VI 30-s) schedules. Inone component, reinforcement consisted ofone count on the subject's displayed tally ofpoints (for which the subject was later paid).In the other component, reinforcement con-sisted of five counts. Thus, the stimuli werecorrelated with different amounts of rewardbut with identical response requirements. Forsubjects to obtain all the scheduled reinforcers,responding on the point-producing lever hadto be maintained at a moderate rate through-out both components. In addition, the mini-mum force required to operate the point leverwas small and approximately the same as thatrequired for the observing levers in this studyand in the studies of Fantino and Case (1983)and Perone and Baron (1980). Hence, shouldthe relative rate of point-lever responding belower in the presence of the less positivediscriminative stimulus (which might be ex-pected because of the differential correlation ofthe discriminative stimuli with amount ofreward), the amount of effort saved by observ-ing would be small. We predicted, therefore,that subjects would more closely replicate thepreferences found by Fantino and Case thanthose found by Perone and Baron- that is,that observing would be consistent with theconditioned-reinforcement hypothesis.

METHOD

SubjectsSubjects recruited were 9 men and 3

women. One subject, who exhibited nearlyzero rates of responding, was dismissed afterthe first session. Data from this subject are notpertinent to the hypotheses under test becauseresponding was unreliable and because stimuliwere produced too infrequently to test eitherhypothesis. Apparently, discovering when the1- versus the 5-point reinforcers were likely tobe available was not reinforcing for this sub-ject. Another subject occasionally defeated theautomatic scheduling circuit by adopting anunusual response topography. Data from thissubject are presented separately from those ofthe remaining 10 subjects and have been ex-cluded from statistical analysis.

Subjects were obtained through a recruit-ment sheet posted in a hallway of the Psy-chology Department. They were paid $2.00per session for their participation plus theamount earned by getting points in the task;point earnings averaged about $2.50 per ses-sion (for a total of approximately $4.50 per ses-sion). Money earned from points was paidafter each session; payment for participationwas made at the end of the experiment.

ApparatusSubjects were seated at a table in a small

room otherwise devoid of furniture. The roomwas sound-insulated and had one door con-taining a small mirrored observation window.Typical indoor environmental conditionsprevailed.A display panel and three levers were on the

table. The display panel was an 80-cm by54-cm metal stand that contained two Sodecocounters (Model TCeBZ4E) and four Syl-vania miniature lamps (Model 28PSB). Onecounter was located on each side of the paneland 10 cm above the table top. The lampswere arranged in a vertical line 2 cm apart andcentered on the panel 10 cm above the tabletop. Lens caps were placed on the bottomthree lamps. From highest to lowest they weregreen, red (or blue), and blue (or red).The levers were identical Microswitch

switches (Model BZ-2RW) equipped with aplastic button 1 cm in diameter. Two hadyellow buttons and were located 30 cm apart,symmetrically in front of the subject. The thirdlever had a green button and was centered be-tween the other levers. The switches requireda force of 1. 2 N to operate and lever travel was1 cm. A lever operation was accompanied byan audible click and illumination of a Sylvaniaminiature lamp that was attached to the levermounting.

ProcedurePoints were obtained according to two ran-

domly alternating VI schedules of reinforce-ment. Points were produced, after a varyingtime averaging 30 s since the last reinforce-ment, by pressing the green-button lever. TheVI schedules differed in the number of points

291

DAVID A. CASE, EDMUND FANTINO, and JOHN WIXTED

per reinforcement. Reinforcement on oneschedule consisted of a single count on one ofthe counters accompanied by a brief flash ofthe green lamp. Reinforcement on the otherschedule consisted of 5 rapid green-lightflashes and 5 counts on the other counter.Component duration (i.e., 1 trial) was 60 s. Ineach of three observing conditions, instruc-tions accurately described the relation ofstimuli to the VI schedules, the function ofpressing the levers, and the significance ofpoints. In one condition- a standard observ-ing condition -the instructions read:

In this experiment you have the chance tooccasionally produce points which are worth1 cent each at the end of today's session. Infront of you are three levers and two coun-ters. Points can be produced only by press-ing the center lever. Some of the time pointswill be recorded on the right counter and atother times they will be recorded on the left.When a successful press is recorded on theright counter, five points will be added at atime. When a successful press is recordedon the left counter, one point will be addedeach time. You have no control over thecounter that records points. Half the time itwill be the right counter and half the time itwill be the left.You can occasionally cause the red or blue

lights to turn on by pressing the two re-maining levers. The red light can be turnedon by pressing the right lever and the bluelight can be turned on by pressing the leftlever. [When the red light is on, it meansthat successful presses will be recorded onthe right (5 point) counter. When the bluelight is on, it means that successful presseswill be recorded on the left (1 point) coun-ter.] If neither light is on you do not knowwhich counter will record a successful press.

Please press the levers with your fingerusing only one hand (your preferred hand).This will insure that only one lever is pressedat a time. You may otherwise select to pressthe levers as you wish in any order. The lightdirectly above each lever will turn on to letyou know you have pressed hard enough.

In two other conditions, the portion of the

instructions in brackets was replaced withother phrases appropriate to the different con-ditions. For an alternative in which the bluestimulus was uncorrelated with reward amount(SU), for example, the instructions in thebrackets read: 'When the blue light is on, itmeans that half the time successful presses willbe recorded on the right counter and half thetime they will be recorded on the left."

In all conditions, presses on the two observ-ing manipulanda (levers with yellow buttonson each side of the point-producing lever) pro-duced stimuli (illumination of the red or bluelamps) in 50% of the trials according to a VI15-s schedule. A stimulus, once produced, re-mained present until the end of the trial. In astandard observing condition, responses onone yellow-button lever illuminated one of thelamps when the 5-point schedule was in effect(i.e., S5 or S+); responses on the other yellow-button lever illuminated the other lamp whenthe 1-point schedule was in effect (i.e., Si orS-). Two other conditions separately testedeach of these observing alternatives relative toa third alternative, in which responses pro-duced a stimulus uncorrelated with amount(i.e., SU). That is, in one of these conditionsresponses on one lever produced S+ when the5-point schedule was in effect while responseson the other lever produced SU. Similarly, thefinal condition arranged for responses on onelever to produce S- when the 1-point schedulewas in effect while responses on the other leverproduced SU.

Subjects served in each of the three observ-ing conditions for one session. The side posi-tion of the alternatives and the color of thelights were counterbalanced and the order ofconditions was varied across subjects. The de-sign is summarized in Table 1.The top, unfiltered lamp on the panel ac-

companied the mixed schedule. It was il-luminated at the start of a session and wasdarkened when the session ended. A sessionconsisted of 32 trials, 16 of each component,prearranged by a stepping switch to occur inan irregular sequence.The sequence of events surrounding a ses-

sion was as follows. After the subject wasseated in the experimental room and the in-

292

RESPONSE EFFICIENCY AND OBSER VING

Table 1Design and Principal Results of Experiment 1

Rate of Observinga and Number ofQuestionnaire Items Answered Incorrectlyb

DesignOrder of Sld Si A:S9(vs.S9) B: 5 (vs. SU) C:A(vs. Su)

Subject Conditionsc Color Side S' SA Errors S')S" Errors SE 5 Errors

1 ACB red left 17 5.8 0 9.5 13 0 12 11 02 CAB blue left 27 12 0 37 33 0 22 6.7 13 BAC red left 3.4 0 0 1.7 13 0 1.3 .2 04 ABC blue right 89 65 0 1.0 3.0 0 1.7 3.6 05 CBA red right 17 0 0 0 .7 0 10 9.8 06 BCA blue right 3.0 2.3 0 .1 5.8 0 7.8 6.2 07 CAB blue left 94 0 0 .6 2.2 1 54 2.9 08 ABC blue right 28 23 1 28 21 0 64 13 09 CBA red right 80 3.3 0 4.3 38 0 70 16 010 BCA blue right 9.7 9.3 0 14 37 0 20 21 3

aExpressed in responses per min.bQuestionnaires contained six items.'The conditions were A = Se (vs. S'); B = S' (vs. SU); and C = S' (vs. SU).dStimulus SI was associated with the 5-point schedule in Condition A and C and was the uncorrelated stimulusin Condition B.

structions were given, the experimenter dem-onstrated the apparatus. Both 5-point and1-point reinforcers were produced and eachobserving lever was used to turn on its cor-

responding light once. (Although one might beconcerned that this demonstration may haveartifically increased the frequency of respond-ing through role playing, it is not plausible toargue that it could have influenced respondingmore on one lever than the other; indeed, ifrole playing were paramount, then no system-atic preferences should have resulted.) Addi-tional instructions were then given to intro-duce the subsequent 10-min trial run in theobserving procedure. A questionnaire was ad-ministered following the trial run in order totest the subject's understanding of the functionof responding and the relation of the lights topoints. Answers to the questions were immedi-ately evaluated. If a question was answered er-

roneously, the subject was carefully correctedbefore proceeding with the full session. Afterthe full session, the same questions and one

additional item were asked. The new questionasked about the subjects strategy of responding.

RESULTS AND DISCUSSION

The error rate was nearly zero in answeringitems on questionnaires administered follow-ing full sessions of observing (Table 1).

Because subjects apparently understood theprocedure and instructions well, lever pressingmay be analyzed with some confidence. Themean overall rate of pressing the center (point-producing) lever was 133 (standard deviation,SD = 68) responses per minute in the mixedschedule and 191 (SD = 79) responses per min-ute in the presence of stimuli produced by ob-serving responses. The mean relative rate ofresponding on the point lever during S+, Su',and S- was calculated with respect to themore positive contingent stimulus in each con-dition. This relative rate, averaged over sub-jects and conditions, was .59 (SD = .18)-thatis, more point-lever responding occurred dur-ing the contingent stimuli correlated with thelarger average amount of reinforcement. Themean absolute rates of responding maintainedduring the S+, SU, and S- stimuli, averagedover subjects and conditions, were 202, 167,and 185 per minute, respectively. It is note-worthy that unequal response rates were main-tained despite the fact that identical variable-interval response contingencies prevailed dur-ing each contingent stimulus and in theirabsence.Mean overall rate of pressing the side levers

in the mixed schedule (i.e., total observingrate summed over levers) was 37 (SD= 36)responses per minute averaged over subjects

293

DAVID A. CASE, EDMUND FANTINO, and JOHN WIXTED

and conditions. The principal finding is that ameasure of choice, the mean relative rate ofobserving averaged over subjects, was consis-tent with the conditioned-reinforcement hy-pothesis in each preference test (individualobserving response rates are presented inTable 1). The results also were inconsistentwith the uncertainty-reduction hypothesis inthe critical comparisons. In each condition,subjects preferred the stimulus that accom-panied the larger average amount of reinforce-ment significantly more, even though the al-ternative stimulus available for observing mayhave been equally or more informative. Theresults show that the mean relative rate ofobserving maintained by the stimulus corre-lated with the 5-point schedule (S+) comparedto a stimulus correlated with the 1-pointschedule (S-), was .76, which is significantlygreater than .50 [t(9) = 3.7, p < .01]. Themore positive discriminative stimulus was pre-ferred even though the stimuli were equally in-formative. In addition, the mean relative rateof observing maintained by the stimulus un-correlated with amount (SU) was .75 whensubjects chose between it and the stimulus cor-related with the 1-point schedule. This wasalso significantly greater than .50 [t(9) = 3.8,p < .01]. That is, an uninformative stimuluswas preferred to a negative informative stim-ulus in the most critical comparison. Finally,the mean relative rate of observing maintainedby S+ compared to Su was .66, againsignificantly greater than .50 [t(9) = 2.4,p < .05]. Although both the conditioned-reinforcement and uncertainty-reduction hy-potheses correctly predicted preference for thestimulus correlated with the 5-point schedulein this latter choice, the total set of results of allthree conditions is consistent only with theconditioned-reinforcement hypothesis.The pattern of results across conditions may

be tested for consistency with the principle oftransitivity in choice patterns within individualsubjects (Fantino & Navarick, 1974; Navarick& Fantino, 1974). Subjects 1, 3, 5, 7, and 9exhibited strong stochastic transitivity in theirchoices. That is, preference for Alternative A(the stimulus correlated with the 5-pointschedule) relative to Alternative C (the

stimulus correlated with the 1-point schedule)was greater than or equal to preferences forAlternative A relative to Alternative B (thestimulus uncorrelated with amount) and Al-ternative B relative to Alternative C. Subjects2, 4, 6, 8, and 10 exhibited morderate sto-chastic transititity in their choices. That is,preference for Alternative A relative to Alter-native C was greater than or equal to only oneof the preferences, A relative to B or B relativeto C. In general, only results that show strongstochastic transitivity can be described by aunidimensional choice theory (Navarick & Fan-tino, 1974). Therefore, two or more variablesmay be required to account adequately for thepattern of preferences in the latter subjects.A strong version of the conditioned-rein-

forcement hypothesis would require absoluteobserving response rates to correlate with theamount of reward obtained during the contin-gent stimuli in choice procedures. Analysis ofabsolute observing rates from this experimentshows that the means were indeed ranked inagreement with this hypothesis. Mean rates ofobserving maintained by the 5-point cor-related, point uncorrelated, and 1-point cor-related stimuli were 31, 13, and 11 responsesper minute, respectively, and a within-subjectsanalysis of variance revealed that the stimulimaintained significantly different rates ofobserving as predicted [F(2, 18) = 4.6,p < .05]. Although the less positive 1-pointdiscrimitive stimulus did not maintainsignificantly less observing than the stimulusuncorrelated with amount, as required by thestrong version of the conditioned-reinforce-ment hypothesis (F < 1), neither was thereany evidence in these data to support theuncertainty-reduction hypothesis.

Subjects gave a variety of answers whenasked for their strategy of responding. Fre-quently answers were a description of whatthey did, not why (e.g., how they distributedtheir responses among the levers). Also, someanswers explicitly claimed no particularstrategy. Other subjects answered that theyeither sought or detected different patterns inthe procedure (e.g., that points were producedin clusters rather than uniformly).

Eleven answers to the strategy question by 6

294

RESPONSE EFFICIENCY AND OBSER VING

subjects were more revealing and relevant tothe hypotheses under test. Answers by 4 ofthese subjects (3, 5, 7, and 9) emphasized therelative rate of getting points associated with astimulus as a reason for preferring a stimulus.For example, one subject wrote, 'The [uncor-related stimulus] told me I could get points onthe [5-point counter] as well as the [1-pointcounter], so I pressed the [corresponding ob-serving] lever more often." Answers by the 2remaining subjects (6 and 8), on the otherhand, emphasized the informativeness of thestimuli. For example, Subject 6 distributedobserving responding as predicted by the con-ditioned-reinforcement hypothesis despite re-porting a strategy that took informativeness ofthe stimuli into account. The subject whoresponded most consistently with the predic-tions of the information hypothesis (Subject 8)wrote, 'I pressed the [observing lever that pro-duced S-] more because the [uncorrelatedstimulus] provided no information." Otherresults for this subject were anomalous: Thehighest average rate of point-lever respondingwas maintained during S- (although the rateduring S+ was higher than during SU).

Certain aspects of the planned contingenciesof reinforcement were reliably circumventedby one subject whose data were excluded fromthe statistical analyses. In particular, for thissubject two successsive presses on the pointlever each produced a point for every onescheduled during the stimulus correlated withthe 1-point schedule, provided the momentaryresponse rate was extremely high just after thefirst point was produced. The subject becameinexplicably skilled at producing the unplannedreinforcer and was much more successful thanthe experimenters ever became during theirattempts to reproduce the effect. Extra pointswere produced at nearly every opportunitythroughout the experiment soon after therapid-burst response pattern developed in thesubject's first session. It is noteworthy that thiswas the only subject who strongly preferredobserving the stimulus reliably correlated withthe 1-point schedule compared to an uninfor-mative stimulus (choice proportion = .77).However, for this subject it is possible that effi-ciency of responding and total earnings could

be detectably increased by observing the lesspositive discriminative stimulus: Point-leverresponding could accelerate during the 1-pointstimulus, thus increasing the chance of pro-ducing an unplanned reinforcer, and couldslow during the other stimulus, allowing thesubject to relax somewhat without endanger-ing loss of scheduled reinforcers. The answersto the strategy questions indicated that theseconsiderations were evident to the subject andthat they controlled responding. It is unclearwhy the contingency was so powerful in shap-ing and maintaining the high rates of point-lever pressing during the 1-point stimulus, butmeeting the inadvertently scheduled high re-sponse rate requirement did mean that the ef-ficiency of point-lever responding could in-crease through producing the stimulus cor-related with the 1-point schedule.

EXPERIMENT 2

The results of Experiment 1 were generallyconsistent with predictions of the conditioned-reinforcement hypothesis regarding preferencein choice procedures, replicating those of Fan-tino and Case (1983). The results suggest thatdiscrepant findings between that study and theone by Perone and Baron (1980) were not dueto presence or absence of a response-reinforcercontingency per se, but could have resultedfrom differences in response efficiency enabledby observing S-. However, many other seem-ingly minor differences between the studies ex-ist that could also be responsible. Thus, theresponse-efficiency argument requires a moredirect test. One such experiment might com-pare within the same subjects the effect ofvarying response efficiency permitted throughobserving. To that end, we attempted in Ex-periment 2 to duplicate the response require-ments in the previous studies in different con-ditions. In one case responding on a 5-lbplunger was required to produce points in thereinforcement component, whereas in the othercase reinforcement was not contingent on anyresponding.. (In this usage, by reinforcementwe mean the same event that under similarcircumstances maintains responding that iscontingent upon it-that is, in situations like

295

DAVID A. CASE, EDMUND FANTINO, and JOHN WIXTED

this people will typically press a lever whenpressing produces points exchangeable formoney.) We predicted that in these circum-stances the rate of observing maintained by Suin the noncontingent condition would begreater than that maintained by S-. On theother hand, we predicted that in the response-dependent condition, preference for S- mightresult depending upon the relative importanceof effort saved versus the negative correlationof S- with points. In any event, if the previousanalysis is correct, preference for S- should in-crease when it enables more efficient respond-ing.

METHOD

SubjectsWe recruited 5 men and 3 women from a

pool of Introductory Psychology students whowere required by their instructor to act as sub-jects for credit towards their course grade.Subjects were paid approximately $2.00 persession in exchange for points obtained in theexperimental task.The data from two subjects, one from each

condition (see Procedure), were excluded fromanalysis. Both ceased responding altogether onthe observing levers in the second session, thusprecluding meaningful interpretation with re-spect to the hypotheses under investigation.

ApparatusThe same room and for the most part the

same apparatus that were used in Experiment1 were used in Experiment 2. One differencewas that this time the display panel containeda single counter that was positioned in thecenter next to the column of stimulus lamps.Also, a plunger (Gerbrands Model G6310)was mounted just beneath the table top andcentered with respect to the display and thelevers. The force required to operate it was 5lbs, the same as that usually used in the ex-periments by Perone and Baron (1980).

ProcedureVI or variable-time (VT) schedules of rein-

forcement randomly alternated with extinction(EXT) within a session, in different conditionsof the experiment. Both VI (or VT) and EXT

schedules were correlated with the top (un-filtered) panel light (inasmuch as it was il-luminated throughout a session) and were ineffect for 60 s before alternating. In theresponse-dependent reinforcement condition,pulls on the plunger illuminated the green(reinforcement available) light according to aVI 60-s schedule. Once the green light was il-luminated, a press on the center lever (con-summatory response) produced a point by in-crementing the counter. The press also dark-ened the green light and started the next inter-reinforcement interval of the VI schedule.Pulls on the plunger during EXT had no effectand the green light always remained dark. Inthe response-independent reinforcement con-dition, the green light was illuminated accord-ing to a VT 60-s schedule independent of anyresponding. Otherwise, the two conditionswere identical. That is, the VT schedule alter-nated with EXT and pressing the center leverin the presence of the green light incrementedthe point counter once and then darkened thelamp. Most other aspects of the procedure(e.g., the schedule by which presses on theobserving levers produced stimuli, session pro-tocol, use of questionnaires, etc.) were thesame as in Experiment 1. However, in orderto simplify the experiment, only one observingcondition from Experiment 1 was used in Ex-periment 2. The condition selected-observ-ing a stimulus correlated with EXT versusobserving one uncorrelated with the sched-ules -is the one for which the predictions ofthe conditioned reinforcement and the infor-mation hypotheses are most discrepant.The instructions for the response-dependent

reinforcement condition read as follows:

[In this experiment you have the chanceto occasionally earn points which are worth10 cents each at the end of today's session.]Points are more likely during certainperiods than during others as you will soonsee for yourself.

In front of you are a plunger, threelevers, and some lights. Points are producedby pulling the plunger. The white light willnormally be on. You can occasionally causethe red or blue lights to turn on by pressing

296

RESPONSE EFFICIENCY AND OBSER VING

the side levers. When the red light is on,some of the time points are twice as likely asnormal. However, at other times when thered light is on, no points can be produced.No points can be produced when the bluelight is on.

Please press the levers with your fingerand pull the plunger using only one hand(your preferred hand). This will insure thatonly one thing is operated at a time. Youmay otherwise select to pull the plunger andpress the levers as you wish in any order.The light directly above each lever will turnon when you press to let you know that youhave pressed hard enough.The green light will turn on when you

produce a point. Press the center lever oncewhen the green light comes on in order toacknowledge that you got it. This press onthe center lever will also cause the greenlight to turn off and the counter in front ofyou to increment. The counter shows thenumber of points that have accumulated sofar. Remember, these points are worth 10cents each.

The instructions for the response-independentreinforcement condition were the same exceptthat all references to the plunger were deletedand the portion in brackets was replaced bythe following:

In this experiment you have the chanceto occasionally receive points which areworth 10 cents each at the end of today'ssession. However, nothing you can do willaffect the total number of points that youreceive- it cannot be either increased ordecreased by your responding.Half the subjects were studied in the two

conditions in an ABA sequence for a total ofthree sessions. The other half were studied in aBAB sequence. Side and color of the observingalternatives were also roughly counterbal-anced across subjects. Table 2 summarizes thedesign.

RESULTS AND DISCUSSIONThe results from questionnaires ad-

ministered after full sessions of observing aresummarized in Table 2. Subjects apparently

understood the procedure well, as evidencedby near-zero error rates in answering items onquestionnaires. Plunger pulling was main-tained well by response-dependent points andcame under the control of the discriminativestimuli. The absolute rate of pulling theplunger in the response-dependent reinforce-ment condition was 82 (SD = 59) responses perminute in the mixed schedule, 110 (SD= 53)responses per minute in the presence of thediscriminative stimulus that was uncorrelatedwith the VI and EXT schedules, and nearlyzero in the presence of the discriminativestimulus correlated with EXT. Negligibleplunger pulling occurred in the response-independent reinforcement condition.

Substantial rates of observing were main-tained in both experimental conditions. Ab-solute rate of pressing the observing levers was39 (SD = 42) responses per minute when pull-ing the plunger produced reinforcement and57 (SD = 59) responses per minute when rein-forcement was independent of responding.The principal results, the relative rate ofresponding on the observing lever that pro-duced S- in each condition, are plotted inFigure 1. The deviation lines on the bars of theresults for individual subjects show the stan-dard deviation of the initial determination andthe replication. Mean preference (relative re-sponse rate) for observing the stimulus cor-related with EXT was .36 (SD = .15) whenreinforcement was presented independent ofresponding. This replicated prior studies inthat it was significantly less than .50 (hatchedbars, 5) = 2.1,p < .05). On the other hand,preference for observing S- increased forevery subject when pulling the plunger pro-duced reinforcement. A within-subjectsanalysis of variance revealed that preferencefor S- was significantly greater in theresponse-dependent reinforcement condition[F(1,5) = 9.8, p < .05]. Although meanpreference reversed in the response-dependentreinforcement condition- that is mean relativerate of observing maintained by S- was .57(SD = .25), the mean was not significantly dif-ferent from equal preference (open bars,t(5) = .63, p < .1). S- was preferred to S' insix of nine replications (p < .1, binomial test).

297

DAVID A. CASE, EDMUND FANTINO, and JOHN WIXTED

Table 2Design and Principal Results of Experiment 2

Rate of Observing' and Number ofQuestionnaire Items Answered Incorrectly'

Design Independent DependentOrder of Sld Si

Subject Conditionsc Color Side Su S- Errors Su S- Errors1 ABA red right 173 3.5 0 77 21 0

Replication 86 18 02 BAB red left 2.4 2.4 0 3.0 6.4 0

Replication 3.4 4.2 03 ABA red right 120 67 0 50 37 0

Replication 28 19 04 BAB blue left 2.1 1.3 1 5.9 10 0

Replication 1.9 1.5 05 ABA blue right 34 11 0 5.2 15 0

Replication 44 24 06 BAB blue left 11 10 1 1.8 12 0

Replication .2 11 0aExpressed in responses per min.'Questionnaires contained 8 items.cThe conditions were A = Independent; B = Dependent.dStimulus SI was uncorrelated with the components.

Responses to the strategy questionquently revealing and indicated th;subjects attempted to use S- to savipulling the plunger in the response-creinforcement condition. Other ans,cated that most subjects avoided ;

response-independent reinforcement

.50

LU 0D

2 z 1.0

LLJ, .50

z 1.0

Pc5= t-tr

E INDEPEMN

DEPENDE

2

n S1

4 5

-m: not -VAl I I -%l-Z IFig. 1. Mean and individual relative r

serving responses maintained by S- in ExpSubjects chose between S- and a stimululated with reinforcement. Only one conditiclicated for each subject, so only the resultcondition for each subject has an error baring error variance of the initial determinati(replication). Error bar length is one standard

l were fre-at severale effort iniependentwers midi-

because no pointswas present.

would be delivered while it

GENERAL DISCUSSION- in the Mean preferences of college students in thiscondition study were consistent with predictions of the

conditioned-reinforcement hypothesis of ob-serving. With few exceptions (notably Subject

OENT 8 in Experiment 1, whose results were anom-alous in other respects) subjects preferred themore positive contingent stimulus with respectto points, except in the response-dependentreinforcement condition of Experiment 2. Inthis condition, production of the less positive

J1n stimulus enabled a substantial improvement inresponse efficiency because subjects could, anddid, reduce ineffective responding on a rel-

6 - atively high-effort manipulandum. Becausethe stimulus correlated with EXT in this casewas also correlated with the saving of consid-erable effort, the preference predicted by the

^ates of ob- conditioned-reinforcement hypothesis cannot)eriment 2. be specified, given the precision of its presentis uncorre- formulation. Both relationships should con-)n was rep- tribute to conditioned reinforcement, but theirpfrom one relative importance is unclear. However, thern and one hypothesis correctly predicted decreased pref-

i deviation. erence for S- for every subject in Experiment 2

298

I

L.*a

RESPONSE EFFICIENCY AND OBSER VING 299

when obeserving no longer could affect the ef-ficiency of responding. That is, the resultsshow that, for the same subjects, S- was notpreferred in the response-independent rein-forcement condition of the experiment. In Ex-periment 1 also, observing did not permit asubstantial improvement in response efficiency.In this case, too, subjects did not prefer anegative informative stimulus. Therefore, theresults show that negative informative stimulimaintained observing only to the extent thatimproved response efficiency was enabledthrough producing them. These data are thusconsistent with Fantino and Case's analysis ofthe ostensibly anomalous findings of Peroneand Baron (1980). These data also argueagainst the hypothesis that uncertainty reduc-tion is a determinant of observing for morehighly evolved organisms. Contrary to theuncertainty-reduction hypothesis, preferencefor informative stimuli depended upon whetherthey were positively or negatively correlatedwith reinforcement and whether or not produc-ing them permitted a more efficient distributionof responding required for reinforcement.

Recent research on observing in childrenlends further support to these conclusions.Mulvaney, Hughes, Jwaideh, and Dinsmoor(1981) studied two normal and two retardedchildren in an observing procedure in whichsubjects could control the duration of exposureto the contingent stimuli. They found that S+was preferred to S- in both groups. Althoughthe difference in duration of exposure declinedover sessions in the normal children (as didtotal observing), a different measure of pref-erence remained clearly consistent with theconditioned-reinforcement hypothesis for bothretarded and normal subjects throughout theentire experiment. Fantino, Case, and Altus(1983) studied normal children of three dif-ferent ages in a procedure identical to thatused in the response-independent reinforce-ment condition of our Experiment 2. Theyfound that the total absolute rate of observingincreased with age but preference for the morepositive contingent stimulus (in one condition,S+ vs. S-, and in another, uncorrelated vs.S-) was constant across ages. In each of twoconditions where differing predictions were

made, the results were consistent with theconditioned-reinforcement hypothesis (prefer-ence for S+ and for the uncorrelated stimulus)and inconsistent with the uncertainty-reductionhypothesis. In addition, they found that pref-erence for the more positive contingent stim-ulus was maintained with extended testing (asit was in Fantino & Case, 1983).The definition of observing behavior spec-

ifies that it be independent of primary rein-forcement. This independence is critical fordissociating contingencies of conditioned rein-forcement from those of delayed primary rein-forcement; it is what distinguishes observingschedules and, for example, chained sched-ules. The reinforcement that the definitionmakes reference to is that provided accordingto the mixed schedule of reinforcement (suchas points in the present study). Another poten-tial reinforcer in response-dependent rein-forcement procedures is the cost or effort in-volved in emitting responses required for rein-forcement. As far as we know, all would agreethat observing which permits increased effi-ciency of responding should be maintainedrelative to a control condition. Therefore, onemight ask why reinforcement of observing bynegative informative stimuli in response-dependent reinforcement procedures hasnever been found in any pigeon study. Oneexplanation is that in such studies the responserequired for reinforcement does not entailsubstantially more effort than the observingresponse. Therefore, the improvement in re-sponse efficiency made possible by observing issmall because the absolute amount of effortsaved is small. This hypothesis is currently be-ing tested directly in our laboratory.

REFERENCES

Berlyne, D. E. (1960). Conflict, arousal, and curiosity.New York: McGraw-Hill.

Bloomfield, T. M. (1972). Reinforcement schedules:Contingency or contiguity? In R. M. Gilbert & J.R. Millenson (Eds.), Reinforcement: Behavioralanalyses (pp. 165-208). New York: Academic Press.

Bowe, C. A., & Dinsmoor, J. A. (1983). Spatial andtemporal relations in conditioned reinforcementand observing behavior. Journal of the ExperimentalAnalysis of Behavior, 39, 227-240.

Case, D. A., & Fantino, E. (1981). The delay-reduc-tion hypothesis of conditioned reinforcement and

300 DAVID A. CASE, EDMUND FANTINO, and JOHN WIXTED

punishment: Observing behavior. Journal of the Ex-perimental Analysis of Behavior, 35, 93-108.

Dinsmoor, J. A., Browne, M. P., & Lawrence, C. E.(1972). A test of the negative discriminativestimulus as a reinforcer of observing. Journal of theExperimental Analysis of Behavior, 18, 79-85.

Fantino, E. (1977). Conditioned reinforcement:Choice and information. In W. K. Honig &J. E.R. Staddon (Eds.), Handbook of operant behavior (pp.313-339). Englewood Cliffs, NJ: Prentice-Hall.

Fantino, E., & Case, D. A. (1983). Humanobserving: Maintained by stimuli correlated withreinforcement but not extinction. Journal of the Ex-perimental Analysis of Behavior, 40, 193-2 10.

Fantino, E., Case, D. A., & Altus, D. (1983).Observing reward-informative and -uninformativestimuli by normal children of different ages. Journalof Experimental Child Psychology, 36, 437-452.

Fantino, E., & Navarick, D. (1974). Recent devel-opments in choice. In G. H. Bower (Ed.), Thepsychology of learning and motivation (Vol. 8, pp.147-185). New York: Academic Press.

Killeen, P., Wald, B., & Cheney, C. D. (1980).Observing behavior and information. PsychologicalRecord, 30, 181-190.

Mulvaney, D. E., Dinsmoor, J. A., Jwaideh, A. R.,& Hughes, L. H. (1974). Punishment of observ-ing by the negative discriminative stimulus. Journalof the Experimental Analysis of Behavior, 21, 37-44.

Mulvaney, D. E., Hughes, L. H., Jwaideh, A. R.,& Dinsmoor, J. A. (1981). Differential produc-tion of positive and negative discriminative stimuliby normal and retarded children. Journal of Ex-perimental Child Psychology, 32, 389-400.

Navarick, D. J., & Fantino, E. (1974). Stochastictransitivity and unidimensional behavior theories.Psychological Review, 81, 426-441.

Perone, M., & Baron, A. (1980). Reinforcement ofhuman observing behavior by a stimulus correlatedwith extinction or increased effort. Journal of the Ex-perimental Analysis of Behavior, 34, 239-261.

Schrier, A. M., Thompson, C. R., & Spector, N. R.(1980). Observing behavior in monkeys (Macacaarctoides): Support for the information hypothesis.Learning and Motivation, 11, 355-365.

Wyckoff, L. B., Jr. (1952). The role of observingresponses in discrimination learning: Part 1.Psychological Review, 59, 431-442.

Received August 15, 1983Final acceptance Februa?y 21, 1985